Selecting the motion ground truth for loose-fitting wearables: benchmarking optical MoCap methods.